Hardest Part

Extracting strict causality and comparative negligence from unstructured, contradictory multi-party narratives like competing driver statements and handwritten police reports with enough accuracy to justify legal recovery costs.

Min Viable Scope

Focus exclusively on personal auto property damage claims in a single regulatory jurisdiction, outputting a prioritized queue of flagged claims with cited evidence for adjusters. Deliberately leave out bodily injury, workers' compensation, and automated outbound legal drafting.

Cold Start Problem

Carriers require proof of high precision before trusting an automated system to flag claims, but precision requires training on thousands of resolved subrogation files. Break this by offering contingency-based historical audits of closed claims for mid-market auto carriers to secure initial data.

Time To First Value

2–4 weeks of historical claim ingestion and batch processing to deliver the first list of missed recovery opportunities

Data Moat Available

true

Technical Difficulty

High

Wedge

Begin with low-dollar ($1,000-$5,000) auto collision claims with clear-liability indicators, such as rear-end accidents documented by police reports. Insurers currently write off this niche entirely, meaning any recovery is pure profit and provides immediate proof of value without disrupting high-stakes adjuster workflows. From there, expand into complex comparative negligence auto claims, and eventually into property damage and workers' compensation subrogation.

Timing

Recent advancements in multimodal LLMs enable reliable extraction of fault attributes from unstructured police reports, handwritten witness statements, and scene photographs. This allows the complete automation of liability determination and demand generation, tasks that previously required human cognition.

Why This ICP

Mid-market auto insurers face acute margin compression and high volumes of minor collisions. Unlike Tier 1 carriers, they lack the scale to negotiate highly optimized offshore BPO contracts, making them immediate adopters for automated recovery systems that stem margin leakage.

Size Of Prize

There are approximately 3,000 P&C insurance carriers and Third-Party Administrators in the US. At an average annual labor spend of $500,000 on subrogation processing and investigation per entity, the addressable labor pool is roughly $1.5B annually.

Gap Narrative

Property and casualty insurers abandon millions of dollars in recoverable claims annually because the cost of human investigation exceeds the claim value. Adjusters manually read police reports, review photos, and draft demand letters, restricting subrogation to high-value incidents. Insurers lack an automated system to parse unstructured evidence and execute recovery demands for the long tail of low-dollar claims.

Defensibility

The product builds a proprietary counter-party knowledge graph over time. As the system issues demands and processes responses, it maps the specific negotiation thresholds, pushback patterns, and settlement behaviors of opposing carriers. This data compounds to increase automated recovery rates and shorten cycle times, creating a distinct outcome advantage that a generic model cannot match.

Why This Thesis

A Service-as-Software approach perfectly aligns with the subrogation problem shape, which is strictly outcome-driven. Insurers do not want another investigation dashboard for their adjusters; they want completed demand packages and recovered funds delivered directly to their ledger.

SAM

~$600-800M US auto and property P&C carriers

SOM

~$30-50M

TAM

~5,000 global P&C insurance carriers × ~$400k/yr in subrogation investigation tooling and labor offset ≈ $2B

Growth Rate

~10-14%/yr, driven by rising claims severity and carrier mandates to lower combined ratios through maximized loss recovery

Paid Comparable Spend

~$60k-90k/yr per human subrogation investigator, plus 15-33% contingency fees paid to outsourced subrogation recovery law firms

Kill Thresholds

• False positive subrogation flag rate > 25% after 30 days of tuning
• Integration time with existing claims management systems > 60 days
• Net new recovered dollars < 2x software cost within the 90-day pilot
• Paid conversion rate from pilot to annual contract < 20%

Leading Metrics

• Subrogation opportunity identification rate per 100 claims
• Time-to-first-flagged-recovery
• False positive rate on fault assignment
• Percentage of AI flags accepted by human recovery teams
• Average dollar value of identified recovery per claim

What Proves Right

Claims adjusters upload loss files and the system identifies missed subrogation targets with a greater than 80 percent accuracy rate compared to human review. Carriers convert trial pilots to $50k-plus annual contracts because the software uncovers at least 3x its cost in previously abandoned recovery dollars. Adjusters route high-confidence subrogation flags directly to recovery teams without manual re-investigation.

What Proves Wrong

Carriers refuse to trust the AI-generated fault assignments due to opaque reasoning, forcing adjusters to manually read the entire claim file anyway. Legal departments block deployment because the tool hallucinates policy limits or local traffic laws during fault determination. Trial cohorts churn after 90 days because the identified subrogation opportunities are too low-value to justify the collection effort.